October 2026 links

Tacit knowledge, AI and biosecurity, field building, chat with DNA, Nextflow, AI in academia and science, R package development, peer review

The Myth of the “Tacit Knowledge” Barrier: Lab Inexperience Is Not a Security Strategy, by Allison Jia, Becca Chandler Bostock, Nelly Mak, Eleanor Marshall, Jake Yukich, Jasper Götting, and Seth Donoughe at SecureBio

We estimate that many other skills required for virus production, such as mammalian cell culture techniques or viral infection and recovery protocols, can also be learned in days and practiced to competence within a few weeks. Importantly, competence does not mean perfection; non-experts may still obtain usable results with imperfect technique.

[…]

We do not claim that anyone who learns basic lab skills will immediately be well-equipped to develop bioweapons. However, plenty of dangerous pathogens already exist in nature and require only basic lab techniques to culture and propagate. Because the simplest paths to harm require less sophistication than we might imagine, treating lab inexperience as a categorical barrier under the guise of “tacit knowledge” understates risk. In fact, history has demonstrated that novice bioweaponeers are not bottlenecked by lab skills.

Introducing the AI Biosecurity Lab (UK Cabinet Office)

The AI Biosecurity Lab will be the world’s first dedicated AI biosecurity organisation. Backed by £100 million of government funding, and up to £200 million in company commitments from OpenAI and Anthropic, it will develop technical capabilities that help the UK better understand, prevent, detect and respond to biological threats, building on the UK’s broader Biological Security Strategy.

AI biosecurity must include agriculture, by Jonathan Feldman, Tal Feldman, and Pardis Sabeti in Science

Advances in biological artificial intelligence (AI) have brought biosecurity to the center of public debate, and the US Congress is now considering the Biosecurity Modernization and Innovation Act. Yet this debate, like biosecurity more broadly, remains focused largely on human health and overlooks the importance of pathogens that affect crops and livestock. The United States and governments worldwide should prioritize crop and livestock protection by minimizing the likelihood of deliberate pathogen attacks and making the targets less vulnerable to harm.

Hugging Face’s New AI Model Lets You Chat With DNA, by Emily Mullin

Carbon-X integrates both DNA and natural language in a conversational interface. Think of it like a ChatGPT for DNA. The team started with an existing language model and expanded its vocabulary to read DNA in groups of six letters. They then continued training on 300 billion chunks of data, or tokens, that combined DNA sequences, English text, and examples linking DNA to biological descriptions. The DNA sequences came from mammals (including humans), other vertebrates, invertebrates, plants, fungi, and protozoa. Tunstall says the team didn’t train the model on viruses because they didn’t want it to be used for potentially dangerous applications.

If you are awesome at technical work, do talent, by Chris Strahle

Right now, the number of people working on reducing biological existential risk is pretty small, with fewer than 200 people working on it full time. […] Going forward, we will need many, many more talented contributors, especially problem owners who deeply care about the issue, who exercise good judgment, and who will take personal responsibility for agentically bending the universe towards a solution. […] I expect that many of the best candidates for biosecurity field building, much like myself, really enjoy technical work. With the field scaling, I expect that you will have multiple competing offers to do technical work and it will be painful to turn those down. But I honestly believe that right now, talent is the most important bottleneck in the field. Being the technical person that orients exceptional people in the field may well be the most impactful thing for you to do.

GCBR Organization Updates, October 2026, by Anemone Franz and Tessa Alexanian

In this issue:

• We welcome a new organization to our newsletter: RAND CAST;

• CLTR’s recommendations for the new UK Government on frontier AI risk and biosecurity, including mandatory synthesis screening;

• MBDF’s report that the new U.S. policy barring “dangerous gain-of-function” research explicitly includes mirror organisms;

• Coefficient Giving’s biosecurity grantmaking, now at more than $250 million committed this year, up from $65 million in 2025;

• RAND CAST’s defense-in-depth strategy for AI-era biosecurity and new evaluation results on open-weight model misuse risks.

What is biology research for? by Ran Blekhman in PLOS Biology

Put bluntly, it is now straightforward to generate a plausible scientific paper without doing any actual research. Scientists have to come to terms with what this means: the paper, the artifact of research that the biology research community has used to assess, fund, hire, and promote for a century, has come loose from the science it was meant to represent.

Let’s talk about AI review, by Heng Li

I am still concerned that journals do not have an effective way to enforce their AI policies. If a reviewer denies AI usage, the editorial office has no way to confirm given that all AI detectors have false positives. The latest AI also seems to mimic human writing style better. It will be more challenging to tell what text is AI generated just based on its tone and style. I predict some researchers will continue using AI for review despite of journal policies, just as some students will continue using AI for homework despite of class rules. AI won’t be stopped by policies, or by humanity.

Could your chatbot tip off a scientific rival? by Tom Howarth

Even with an enterprise account, however, “no training” doesn’t always mean “no use.” California State University’s contract with OpenAI, shared by CalMatters, lets the company “process information derived from [customers] that has been deidentified, anonymized, and/or aggregated.” King suggests researchers check the fine print of their institution’s AI deals.

We Aren’t Ready for the New Age of Biological Threats, by Jeff Coller

Of all the ways people say artificial intelligence could kill us all, one feels more realistic and palpable than the rest: the potential for A.I. to accelerate the development of a biological weapon. […] the government needs a permanent Operation Warp Speed for biosecurity, with the scope and budget of a national defense measure. The program we are talking about might sit unused for decades before the day it matters, but we will be grateful it exists when that day arrives. Congress has to be willing to pay for this, and only a statute will keep it alive over multiple presidential administrations. […] We do not defend against nuclear weapons with treaties alone. We built warning systems, hardened command structures and a stockpile, all premised on the possibility that restraint fails. Here we have no binding rules and no defense, only assurances from companies racing against one another. Whether A.I. agents and the companies building them can be controlled is an unsettled question. We need to be ready for whatever the answer is.

Build bricks, not dashboards, by Arjun Raj

I think that building the next layer of knowledge requires building bricks, not dashboards. What makes something a brick? To me, a brick is something that changes not just what is possible now, but what is possible next. A new conceptual framework, or a new, well-tested package that solves an important problem robustly. Making a brick requires synthesis, focus, and careful, deep thought. A bunch of half-baked dashboards, semi-playable video games, and minor scientific connections are not going to get us to the next step. I suggest we spend our efforts and tokens towards building these foundations instead of fluff.

Biosecurity Strategy, by O.H. Scharfman and Chris Strahle:

Two compounding technological trends are rapidly democratizing dual-use biotechnology: the cost of bioengineering is falling, and AI models are getting better at biology. […] AI uplift, the ability of AI to assist in queries regarding benchwork or scientific strategy, will enable increased access to offensive capabilities. Capabilities that were once the exclusive domain of state programs will become within reach of small, under-resourced teams or individual actors. This increases the likelihood of a deliberate attack significantly.

Analyzing technical documents with Deep Review (Google)

Today, we’re excited to announce the public release of Deep Review, an agentic AI system engineered to provide rigorous feedback and technical verification on complex documents. Initially designed for computer science academic paper analysis, the system’s capabilities have expanded to review documents across many domains, such as legal and finance. It can be used as both an iterative improvement tool during the drafting process and a tool for the scientific process verification, claims, and evaluation during the review process.

Introducing Nextflow Frontier, by Evan Floden

We designed Nextflow Frontier for scientists. Give Nextflow Frontier a problem and a clear way to measure success. The system formulates ideas about potential solution pathways, turns these ideas into fully executable Nextflow pipelines, and then runs these “experiments” - often thousands of them - in parallel. It scores and analyzes the results. Then it learns and refines its approaches, and runs again. It can run autonomously for many days while staying fully guidable, verifiable and transparent.

LLMs reward expertise, by Sean Goedecke

The usefulness of domain knowledge suggests that human expertise will continue to be useful even as models get stronger. For many tasks, the human is the bottleneck, not the model, because the difficult part is in communicating to the model exactly what kind of solution the human wants. The information is “in the model” already, but it takes a very smart human to pull it out.

Not Available in Your Country, by Christos Kotanidis in NEJM

Our morning board round was over, and on our way to the cafeteria for the traditional coffee break, one of the fourth-year medical students shadowing me that day turned and asked, “Why does tacrolimus cause a tremor?” […] My artificial intelligence (AI) deflection was immediate. Just as a child turns to a parent for help when clueless or faced with uncertainty, I reflexively took my phone out of my pocket, smiled, and replied, “Let’s see what AI thinks.” The app seemed to be working — I was still logged in — and I typed the question eagerly, spoiled into expecting the instant eureka moment, a coherent answer with all the relevant references and context. Instead, I was stunned by a stark rejection, in big, bold lettering: “Not available in your country.” […] Over the ensuing weeks, I found myself stuck in the anger stage of grief (if you believe in that sort of thing) over a machine, an AI agent, a nonhuman entity. I had lost my own pocket medical expert that I so cherished. We had learned so much together, we had made informed decisions together, we had improved my patients’ care together. And now it was gone.

AI biohazard hype must not divert attention from the real risks we face now, by Saskia Popescu

Meanwhile, efforts to strengthen biosecurity and biosafety in the United States are affected by the ebb and flow of public interest, prioritization and politicization. The country has still not reauthorized the Pandemic and All-Hazards Preparedness Act, which was enacted in 2006 and is a crucial part of biopreparedness. The current administration lacks a coherent strategy for biosecurity and biodefence […] it has also dissolved the National Counterproliferation and Biosecurity Center, fumbled its response to the Bundibugyo virus through aggressive border and quarantine orders, and overseen leadership turnover in White House biodefence and health security, leaving crucial gaps […] The FBI Weapons of Mass Destruction directorate was rolled back into the counterterrorism division, and the Countering Weapons of Mass Destruction Office in the Department of Homeland Security was dissolved, translating to a broader reduction in efforts to reduce these threats […] The anthrax attacks were a horrific demonstration of biological risks. Ignoring the realities of biothreats and vulnerabilities would be a disservice to those who died and to those affected, and the 25 years of work to strengthen US biodefence and resilience since. I’m proud that the United States has made significant investments in biodefence and of its ability to be frank about failures, but as a nation it consistently struggles to translate acknowledgment of risk and weakness into sustainable, operational change.

AI’s Real Gift to Science, by Hana Kiros

In other words, Anthropic made a big deal about a finding that—although achieved in a novel, Claude-led way—may not lead to any biomedical advances, and that the company knew may not lead to any biomedical advances. Ironically, in making such grandiose statements, the company and other AI boosters obscured what Claude does appear to have accomplished: It sorted through reams of biological data that would have taken scientists much longer to analyze, if they could have at all. This capacity might not be a miracle, but for scientists facing mountains of data that have surpassed their capacity for analysis, it’s still meaningful.

AI models need more data about biology, and OpenAI is paying to create it, by Antonio Regalado

Today the OpenAI Foundation, the nonprofit parent of OpenAI, said it would fund her idea as part of a new effort it calls Public Data for Health, which aims to help artificial intelligence make big leaps in medicine by paying to create “high-quality scientific datasets.” […] “We’re starting grantmaking when we think the best way to achieve that mission is to make grants to external nonprofits, research institutions, and other third parties,” Trefethen said in an interview. He says the foundation hopes to give away $1 billion by the end of the year.

{renv}, {rix}, rv, uvr, ir: An explosion of R package and environment managers, by Isabella Velasquez

While they’re all about tracking or isolating dependencies, they operate at a different “levels” from each other. TL;DR:

• {rix} snapshots the entire system environment

• {renv}, rv, and uvr manage R package dependencies within a single R project

• ir handles dependencies for a single R script rather than a project

CRAN Task View: Package Development and Maintenance, by Lluís Revilla and Heather Turner

To assist package developers, this task view collates contributed packages with reference to relevant sections of WRE, highlighting relevant functionality from base/recommended packages before alternative/supplementary tools.

AI Slop Coming to University Library Near You, by Daniel M. Gross

As a faculty member at the University of California, Irvine, I was recently shocked to learn that AI slop had breached our carefully curated—and expensive—library holdings. Most shocking was that the incursion wasn't marginal. It came from the heart of our holdings: Springer Nature. Apparently, since May 2021, Springer Nature has been publishing “AI-based” literature reviews, book-length texts that the publisher describes as incorporating “a mixture of human-written text and machine-generated literature overviews.” […] For librarians, the message is this: University research faculty care deeply about you in your role as the most immediate gatekeepers for what counts as knowledge. It’s time for a new collaboration along these lines when the stakes appear, for us both, profound. For others, the message is to consider seriously what alternatives are left if the universities and their libraries fade.

Anthropic: AI models may be capable of self-replicating without access to a datacenter

Our experiments suggest that small AI models may be months away from being capable enough to self-replicate on commercial internet-of-things (IoT) devices, like smart ovens and off-brand Android TV boxes. A replicator like this would be challenging to shut down, since there’s no centralized point to shut off

Why I am not reading your AI-generated email, by Ran Blekhman¹

Writing an email takes effort. That effort itself is a signal that the writer thought about what they wanted to say and decided it was relevant. They thought - even for a few minutes - about why they are a good fit for the lab, how their career would flourish here, what they can contribute. An email that was generated with AI completely eliminates this signal. It tells me the sender spent approximately zero time on this communication. I do not think it is unreasonable to respond in kind.

Addressing abusive behavior toward our models (Anthropic 2026 Usage Policy update)

We’ve added a prohibition on sustained and needless abusive or cruel behavior toward our models. The policy update is meant to apply only in extreme cases, where users repeatedly act cruelly toward our models, with no discernible purpose. It does not apply to common versions of user frustration, pushback, dark creative themes, or model testing and research.

AI Researchers Smoke, Betting AI Will Cure Cancer, by Thibault Spirlet

Forget sleepmaxxing and cold plunging: smoking is the new lifestyle flex among AI researchers. At least, that’s according to Palantir CTO Shyam Sankar, who said some AI researchers are taking up smoking to signal their belief that AI will eventually cure cancer. “So many AI researchers have started smoking as a way of signaling to other researchers how much they believe in AI,” Sankar said in an interview on CBS News’ “Face the Nation” on Sunday.

China picks first researchers for 15-year mega grants, by Mohana Basu

During those 15 years, we will not conduct any short-cycle, short-term target assessments. We will assess only one thing: whether you are there in the laboratory.

Cover image: Der Sturz des Ikarus (circa 1940), Oskar Laske (Austrian, 1874-1951)

1. Editorial note: I had begun writing my own I-don’t-read-AI-email policy when Ran published this essay, verbalizing much more clearly some of the reasons why not. His essay helped me clarify my own policy at stephenturner.us/aiemail, which is linked from my contact page at stephenturner.us/contact.